Berkeley
Current/Voltage Commutating Mixers
Prof. Ali M. Niknejad
U.C. BerkeleyCopyright c© 2015 by Ali M. Niknejad
1 / 50
Current Commutating Mixers
+LO −LO
VCC
Q2 Q3
Q1+RF
IF
RIF
+LO −LO
+RF
IF
A popular alternative mixer topology uses a differential pairLO drive and an RF current injection at the tail. In practice,the tail current source is implemented as a transconductor.
The LO signal is large enough to completely steer the RFcurrent either through Q1 or Q2.
2 / 50
Current Commutating Mixer Model
+LO −LO
VCC
Q1+RF
IF
RIF
If we model the circuit with ideal elements, we see that thecurrent IC1 is either switched to the output or to supply at therate of the LO signal.
When the LO signal is positive, we have a cascode dumpingits current into the supply. When the LO signal is negative,though, we have a cascode amplifier driving the output.
3 / 50
Conversion Gain
tTLO
+1
0
s(t)
We can now see that the output current is given by a periodictime varying transconductance
io = gm(t)vs = gmQs(t)vs
where s(t) is a square pulse waveform (ideally) switchingbetween 1 and 0 at the rate of the LO signal. A Fourierdecomposition yields
io = gmQvs
(0.5 +
2
πcosω0t −
2
π
1
3cos 3ω0t + · · ·
)
4 / 50
Conversion Gain (cont)
So the RF signal vs is amplified (feed-thru) by the DC termand mixed by all the harmonics
ioVs
=gmQ
2
(1
2cosωst +
2
πcos(ω0 ± ωs)t − 2
3πcos(3ω0 ± ωs)t + · · ·
)
The primary conversion gain is gc = 1πgmQ .
Since the role of Q1 (or M1) is to simply create an RFcurrent, it can be degenerated to improve the linearity of themixer. Inductance degeneration can be employed to alsoachieve an impedance match.
MOS version acts in a similar way but the conversion gain islower (lower gm) and it requires a larger LO drive.
5 / 50
Differential Output
+LO −LO
+RF
IF
+LO −LO
+RF
+IF −
This block is commonly known as the Gilbert Cell
If we take the output signal differentially, then the outputcurrent is given by
io = gm(t)vs = gmQs2(t)vs
6 / 50
Differential Output Gain
t
TLO
+1
−1
s2(t)
The pulse waveform s2(t) now switches between ±1, and thushas a zero DC value
s2(t) =4
πcosω0t −
1
3
4
πcos 3ω0t + · · ·
The lack of the DC term means that there is ideally no RFfeedthrough to the IF port. The conversion gain is doubledsince we take a differential output gc/gmQ = 2
π
7 / 50
Double Balanced Mixer
IEE
+LO
−LO
VCC
Q1 Q2 Q3 Q4
Q5 Q6
+LO
+RF −RF
RE
IF
RIF
RE
The single balancedGilbert cell withdifferential output rejectsthe RF at the IF port butthe LO feed-throughremains.
A double balanced mixerhas a differential RF anda differential LO (doublebalanced)
8 / 50
Double Balanced Mixer LO Rejection
As before, the RF stage is a transconductance stage.Degeneration can be used to linearize this stage.
Because the bias current at the output is constant IEE/2regardless of the LO voltage, the LO signal is rejected. Thisrelies on good matching between transistor Q2 and Q4.
The differential operation also rejects even order distortion.Viewed as two parallel Gilbert cells, this mixer is also morelinear as it processes only half of the signal.
The noise of this mixer, though, is higher since the noise ineach transistor is independent.
9 / 50
Mixer Operation
IEE
+LO
−LO
VCC
Q1 Q2 Q3 Q4
Q5 Q6
+LO
+RF −RF
RE
IF
RIF
RE
+is
−is
IEE
+LO
−LO
VCC
Q1 Q2 Q3 Q4
Q5 Q6
+LO
+RF −RF
RE
IF
RIF
RE
+is−is
The AC operation nearly identical to a single balanced Gilbertcell.
Even a single ended output, though, effectively multiplies thesignal by ±1, thus rejecting the RF signal.
10 / 50
Mixer Noise Folding
Since the mixer will downconvert any energy at a distance ofIF from the LO and it’s harmonics, all the noise from theseimage bands will be downconverted to the same IF.
For a Gilbert cell type mixer, the current of the Gm stageproduces white noise which is downconverted. Summing overall the harmonics, we have
∞∑k=−∞
|ck |2 =1
T
∫ T
0s(t)2dt
where the last equality follows from Parseval’s Theorem.
For a square waveform s(t), the harmonic powers fall like1/k2. Thus the third harmonic contributes about 10% to theswitching pair noise.
11 / 50
Mixer Noise Due to Switching Pair
Q1 Q2
In addition to the noisefolding discussed above,the switching pair itselfwill contribute noise at IF.If Q1/Q2 are both on,they act as a differentialamplifier and introducenoise.
When only Q1 or Q2 is on, though, the noise is rejected dueto the degeneration provided by the transconductance device.
Thus we generally use a large LO signal to minimize the timewhen both devices are conducting. It’s therefore not surprisingthat the noise figure of the mixer improves with increasing LOamplitude.
12 / 50
Simulating Mixers with SPICE
Let’s take a typical example of RF = 1GHz, IF = 1MHz.This requires an LO = RF ± IF , close to the RF.
To resolve the IF frequency components, we need to simulateseveral cycles of the IF, say 10, so Tsim = 10TIF
But that means that we must simulate 10, 000 RF cycles
Tsim
TRF=
10TIF
TRF=
10fRFfIF
=10 · 1000
1= 10, 000
If we are conservative, we may insist that we simulate at least10 points of the RF cycle, that implies 100, 000 points persimulation.
This is a long and memory intensive simulation. Weencountered a similar problem when simulating IM3
13 / 50
Periodic Steady-State (PSS) Simulation
Transient simulation is slow and costly because we have to doa tight tolerance simulation of several IF cycles with a weakRF.
The SpectreRF PSS analysis is a tool for finding the periodicsteady-state solution to a circuit. In essence, it tries to findthe initial condition or state for the circuit (capacitor voltages,inductor currents) such that the circuit is in periodic steadystate.
It can usually find the periodic solution within 4-5 iterations.
In the mixer, if we ignore the RF signal, then the periodicoperating point is determined by the LO signal alone.
14 / 50
PSS Iteration
Since typical PSS run converges in 4-5 cycles of the LO, or asimulation time of about 5TLO , the overall simulationconverges several orders of magnitude faster than transient atthe IF frequency.
PSS requires that the circuit is not chaotic (periodic inputleads to a periodic output).
High Q circuits do not pose a problem to PSS simulationsince we are finding the steady-state solution. The high Qnatural response takes roughly Q cycles to die down, thussaving much simulation time.
15 / 50
PSS Options
We can perform PSS analysis on driven or autonomouscircuits. An autonomous circuit has no periodic inputs butproduces a periodic output (e.g. an oscillator).
For PSS analysis we need to specify a list of “large” signals inthe circuit. In a mixer, the only large tone is the LO, so thereis only one signal to list.
We also specify the “beat frequency” or the frequency of theresulting periodic operating point. For instance if we drive acircuit with two large tones at f1 and f2, the beat frequency is|f1 − f2|. Spectre can auto-calculate this frequency.
An additional time for stabilization tstab can be specified tohelp with the convergence. For a mixer this is not needed.
16 / 50
Periodic AC (PAC)
Once a PSS analysis is performed at the “beat frequency”,the circuit can be linearized about this time varying operatingpoint.
Note that for a given AC input, there are as many transferfunctions as there are harmonics in the LO.
We specify the frequency range of the AC input signal aseither an absolute or relative range. A relative range is afrequency offset with respect to the beat frequency.
We also specify the maximum sidebands to keep for thesimulation. Note that this does not affect the accuracy of thesimulation but simply the amount of saved data.
17 / 50
PAC Picture
1.0
0.9
0.12.0
3.01.9
2.91.1 2.1 3.1
fo 2fo 3fofifo − fi
2fo − fi 3fo − fi 4fo fi−fo + fi 2fo + fi
The input frequency fi is translated to frequencies fi + kfo ,where fo is the beat (LO) frequency. The k = 0 sidebandcorresponds to the DC component of the LO signal (e.g. timeinvariant behavior). The non-zero components, though,correspond to mixing. E.g. k = −1 correspond to frequencydown-conversion. k = +1 is the normal up-conversion.k = −2 is the 2nd harmonic mixing fi − 2fo .
18 / 50
PNoise
fo 2fo 3fofifo − fi
2fo − fi 3fo − fi 4fo fi−fo + fi 2fo + fi
PNoise is a noise analysis that takes the frequency translationeffects into account. The simulation parameters are similar toPAC with the exception that we must identify the input andoutput ports (for noise figure) and the reference side-band, orthe desired output frequency. For a mixer, this is k = −1.
19 / 50
Voltage Switching Mixers
LO
LO
+RF
−RF
IF
RIF
Instead of switching currents, we can also switch the voltage.
In the above circuit, during the +LO cycle, switch S1activates and feeds the RF to the output directly. In the −LOcycle, switch S2 activates and feeds an inverted RF signal tothe output.
This circuit requires good switches that turn on hard (lowon-resistance) and turn off well (good isolation).
20 / 50
MOS Switching Mixer
LO
LO
+RF
−RF
IF
RIF
A practicalimplementation uses MOSdevices as switches. Thedevices are large tominimize theiron-resistance with a limitdetermined by isolation(feed-throughcapacitance).
We see that the RF signalis effectively multiplied by±1 with a rate determinedby the LO signal.
A differential RF signal is created using a balun or fed directlyfrom a balanced LNA.
21 / 50
MOS Device Feedthrough
Cgs Cgd
CjdCjs
Rbs Rbd
RchCjdCjs
Rbs Rbd
Cgso Cgdo
Cgb
Rbg
G
S D
B
G
S D
B
When the device is “on”, it’s in the triode region. Due to thelow on-resistance, the coupling through the substrate and LOpath is minimal.
When the device is “off”, the RF and LO leak into the IFthrough the overlap and substrate capacitances.
22 / 50
Summary of MOS Switching Mixer
MOS passive mixer is very linear. The device is either “on” or“off” and does not impact the linearity too much. Since thereis no transconductance stage, the linearity is very good.
The downside is that the MOS mixer is passive, or lossy.There is no power gain in the device.
Need large LO drive to turn devices on/off
Need to create a differential RF and LO signal. This can bedone using baluns or by using a differential LNA and LObuffer.
23 / 50
MOS “Ring” Mixer
+RF
−RF
LO
LO
LO
LO
IF
The RF/LO/IF are all differential signals. During the positiveLO cycle, the RF is coupled to the IF port with positive phase,whereas during the negative phase the RF is inverted at the IF.
The MOS resistance forms a voltage divider with the sourceand load and attenuates the signal as before.
24 / 50
Passive Mixer LO Power
LO
RF IF
LO
RF IF
Since gates of the MOS switches present a large capacitiveload, a buffer is needed to drive them.
The LO buffer can be realized using larger inverters (approach“square wave”) or as a tuned buffer. A tuned lowers thepower by roughly Q but has a sinusoidal waveform .
25 / 50
LO Power (cont)
For an inverter chain driving the LO port, the powerdissipation of the last stage is given by
Pinv = CV 2LO fLO
C is the total load presented to the LO (two MOS devices forthe double balanced mixer, one MOS device for singlebalanced).
VLO is the LO amplitude to fully turn the devices on and off.The devices should be biased near threshold. fLO is the LOfrequency.
26 / 50
Tuned LO Power
For the tuned load case, the power is given by
Ptuned =V 2LO
2Rt
Rt is the effective shunt resistance of the tank. Since the tankQ = ωLORtC , we have
Ptuned =V 2LO
2QωLOC =
πCV 2LO fLO
Q
A high Q tank helps to reduce the power consumption of theLO buffer.
27 / 50
H-Bridge Ring Mixer
+RF
−RF
IF LOLO
If PMOS devices are available, two CMOS inverters form anH-Bridge, applying the RF input signal to the IF directlyduring the LO cycle and inverting the RF input at the IFoutput in the LO cycle.
PMOS and NMOS devices are sized appropriately to maximizeon-conductance and to minimize off capacitance.
28 / 50
Time-Varying Conductance
+ IF −
+RF
−RF
g(t)
g(t − TLO/2)g(t)
g(t − TLO/2)
The RF voltage is appliedto a time varyingconductance. Note that ifthe conductance of a theLO switches is given byg(t), then theconductance of the LOswitches is given byg(t − TLO/2).
The Thevinin equivalent source voltage is given by the opencircuit voltage
vT = vRF
(g(t)
g(t) + g(t − TLO/2)− g(t − TLO/2)
g(t) + g(t − TLO/2)
)
vT = vRF
(g(t)− g(t − TLO/2)
g(t) + g(t − TLO/2)
)= m(t)vRF
29 / 50
Time-Varying Gain m(t)
For the MOS device and a given LO waveform, the functionm(t) can be calculated and the Fourier expansion can be usedto derive the gain.
In parctice there is a load capacitance CIF at the IF port tofilter the downconverted signal. This CL complicates theanalysis but interested students are encouraged to read thepaper by Shahani, Shaeffer and Lee (JSSC vol. 32, Dec 1997,p. 2061-1071)
30 / 50
“Passive” Current Mixer+LO
−LO
+IF
−IF+RF −RF
+LO
The input stage is a Gm stage similar to a Gilbert cell mixer.The Gilbert Quad, though, has no DC current and switcheson/off similar to a passive mixer.
The output signal drives the virtual ground of a differentialop-amp. The current signal is converted into a voltage outputby gthe op-amp.
31 / 50
Advantages of “Passive” Current Mixer
No DC current in quad implies that there is no flicker noisegenerated by the switching quad. This is the key advantage.
The linearity is very good since the output signal is a current.The voltage swing does not limit the linearity of the mixer.This is to be contrasted to a Gilbert cell mixer where thevoltage swing is limited due to the headroom of the switchingmixer and the transconductance stage.
The op-amp output stage can be converted into an IF filter(discussed later)
32 / 50
Disadvantages of Mixer
Need large LO drive compared to the active Gilbert cell mixer.
Need an op-amp. This requires extra power consumption andintroduces additional noise.
Need a common mode feedback circuit at the input of theop-amp.
33 / 50
Ring or Quad?
+LO
−LO
+LO
M1
M4
M3
M2
+LO
M1
−LO
M2
+LO
M4
−LO
M3+RF
−RF
+RF −RF+IF
−IF
+IF
−IF
Note that the Gilbert quad is really a folded ring. Thus thepassive and active mixers are very similar. The main differenceis how the quad devices are biased. In the Gilbert cell they arebiased nominally in saturation and have DC current. In thepassive mixers, they are biased near the threshold.
34 / 50
Op-Amp Noise+LO
−LO
+IF
−IF
+LO
Cp
Rp
Cp
The op-amp input referred noise is amplified to IF. Theresistance seen at the op-amp input terminals is actually aswitched capacitor resistor!
The parasitic capacitance at the output of thetransconductance stage is charged and discharged at the rateof the LO.
35 / 50
Switched Capacitor Resistor+LO
−LO
Cp
Ix
+Vx
−
x
+V−
Note that the parasitic capacitances are charged at the rate ofthe LO to the input voltage Vx , and then to the −Vx , everycycle.
The total charge transferred during a period is given by
Qtot = CpVx − (−CpVx) = 2CpVx
The net current is given by Ix = QtotTLO
= 2CpVx fLO
36 / 50
Switched Capacitor Resistor (cont)
Since there are two differential pairs connected to the op-ampterminals in parallel, the total charge is twice. So the effectiveresistance seen at this node is given by
Rp =Vx
2Ix
The effective resistance is therefore given by
Rp =1
4fLOCp
This is a switched capacitor “resistors”.
37 / 50
Op-Amp Noise Transfer
+IF
−IF
Rp
Rp
Rf
Rf
The noise is thus transferred to the output with transferfunction given by
v2o =
(1 +
2Rf
Rp
)2
v2amp
To minimize this noise, we have to minimize the parasiticcapacitance Cp and the op-amp noise.
38 / 50
Output Filter Stage
RF
LO
IF Filter
+IF
−IF
Since a down-conversion mixer will naturally drive a filter, wesee that the output current can be used directly to drive acurrent mode filter.
For instance, the op-amp can be absorbed into the first stageof a multi-stage op-amp RC IF filter. The feedback resistor Rf
is shunted with a capacitor Cf to produce a pole.
39 / 50
N-Path Filters
Sometimes it’s easier to build a high-Q low-pass filter than abandpass filter, especially on-chip. We can use mixers tosynthesize a bandpass characteristic by simplydown-converting the signal to baseband, performing thefiltering using a low-pass filter, and then up-converting again.
Notice that from the signal path perspective, it’s as if weapplied a bandpass filter.
40 / 50
Bi-Directional Mixers
+LO
−LO
+RF
+LO
−RF
+BB
−BB
Zin(ω)
This is just an N-path filterbut one mixer performs bothup-conversion anddown-conversion. In thiscase, we can synthesize ahigh Q input impedancewithout high Q resonatorsusing a large basebandcapacitor.
This works because the MOS device is bi-directional.
Notice that signals flow in both directions. If an RF signal isdriven at the input and filtered at the output, then the outputsignal is further up-converted and loads the RF port at the RFfrequency !
41 / 50
Diode Mixers
Bias
RF
IF
LO RIFC0
L∞
If MOS devices are not available, or if the frequency exceedsthe capability of the MOS devices, diodes can be used asswitch elements.
Here the diode is biased into “on” state by the chokes andbypass capacitors. A strong LO signal, though, can turn thediode off.
42 / 50
Weak LO Diode Mixer
RF
IF
LO RF IF
RFLO +
The RF signal “riding” on top of the LO signal is thus mixedby the switching action.
Note that if the LO signal is weak, then mixing action occursby the non-linearity of the diode. But this mode has a smallerconversion gain and produces more distortion.
43 / 50
Diode Ring Mixer
RF
IF
LO
RIF
D1
D4
D3
D2
The ring mixer is a passive mixer. Typical loss numbers areabout 6dB. The mixer has good isolation. The LO ←→ RFisolation comes from the transformers. The balancedoperation results in RF and LO rejection.
The LO signal drive is large since it must turn on/off diodes.
44 / 50
Mixer Description
The ring mixer is a fully balanced mixer but we have theoption of driving the LO and RF single ended. As we shallsee, the LO signal alternatively turns on diodes D1/D2 orD3/D4 thereby connecting the RF voltage to the IF withalternatively polarity.
During a positive LO voltage, the secondary of the LOtransformer applies a positive voltage across D1 in series withthe load R thereby forward biasing the diode.
Likewise, the secondary terminal of the LO transformer appliesa negative voltage to the cathode of D2, thereby forwardbiasing it in series with the load.
Diodes D3/D4 are reverse biased and therefore open circuits(ideally).
45 / 50
Positive LO Cycle
RF
IF
LO
RIF
Zi = 0
+
vs
−
+vs
N−
The equivalent circuit above shows small resistor representingthe diode on resistance.
The RF signal is applied to the load through the center tap ofthe transformer. We shall show that the impedance lookinginto the transformer is low (ideally zero), and thus the entireRF signal at the secondary terminal is applied to the loadvo = −vs/N
46 / 50
LO Transformer
+vx
−
Zi =vx
ix
i1
i2
ZLO
Note the input impedance looking into the LO transformer isideally short circuit. To see this, apply a test voltage vx tothis node.
If we ignore the effect of the LO signal for now, we see thattwo equal currents i1 = i2 = ix/2 flow into the secondary ofthe LO transformer. The return current comes from thecenter tap ground.
47 / 50
LO Transformer (cont)
The induced voltage at the primary is given by
v1 = jω(M12 −M13)ix/2 ≡ 0
This is because M13 = M12 and the currents on the secondaryare out of phase.
The voltages induced on the secondary side are likewise zero(assuming perfect coupling)
v2 = jωL2ix/2 + jωM23(−ix/2) ≡ 0
v3 = jωL3(−ix/2) + jωM32ix/2 ≡ 0
48 / 50
Negative LO Cycle
RF
IFRIF
Zi = 0
+
vs
−
+vs
N−
LO
By symmetry, during the negative LO cycle, the mixersimplifies to the above equivalent circuit.
During this cycle, the RF signal is applied to the load throughthe center tap and the bottom of the transformer, thusproducing an output signal in phase with the RF
vo = +vs/N
49 / 50
Diode Ring Mixer Operation
The operation is thus similar to the double balanced Gilbertcell mixer in voltage mode. The RF voltage is multiplied by±1 with a rate of the LO signal.
The lack of DC means that the RF is rejected at the IF port.Likewise, the LO/RF signals are isolated by the switches. Theonly feed-through occurs due to reverse isolation of the diodes.
The biggest drawback, though, is the bulky transformers andthe large LO drive. The mixer is very linear and quiteattractive in applications where linearity reigns supreme overpower consumption.
At microwave frequencies, the transformers can be replaced bycouplers. Since diodes can operate up to extremely highfrequencies (THz), the entire circuit can work up to THzregion.
50 / 50